bridgedatav2_train

bridgedatav2_train数据集是一个从Hugging Face数据集ZibinDong/bridgedatav2_train转换而来的Apache TsFile格式数据集，专注于机器人学和时间序列数据。该数据集基于BridgeDataV2源数据，采用LeRobot Dataset V3.0格式，原始数据来自伯克利RAIL实验室的BridgeDataV2项目。源数据集规模庞大，包含61,529个episodes、2,228,465帧图像、11,950个任务，数据采集帧率为5 fps，使用WidowX250机器人，视频流为图像模态。转换后的数据集以时间序列形式组织，包含三个.tsfile数据文件，以及元数据文件如info.json、stats.json等，但视频数据未包含在此仓库中，仍保留在原始Hugging Face数据集中。数据模式包括毫秒级时间戳、标签列（如episode_index和task_index）和字段列（如frame_index、state_0至state_6、action_0至action_6等），其中向量列被展平存储为单精度浮点数。该数据集适用于机器人控制、时间序列分析和机器学习任务，用户可以使用Apache TsFile SDK读取数据，并通过SQL-like查询过滤特定episode。The bridgedatav2_train dataset is an Apache TsFile format dataset converted from the Hugging Face dataset ZibinDong/bridgedatav2_train, focusing on robotics and time series data. It is based on the BridgeDataV2 source data, using the LeRobot Dataset V3.0 format, with original data from the BridgeDataV2 project by the Berkeley RAIL lab. The source dataset is large-scale, containing 61,529 episodes, 2,228,465 image frames, and 11,950 tasks, collected at a frame rate of 5 fps using a WidowX250 robot, with video streams in image modality. The converted dataset is organized in time series format, including three .tsfile data files and metadata files such as info.json and stats.json, but video data is not included in this repository and remains in the original Hugging Face dataset. The data schema includes millisecond timestamps, label columns (e.g., episode_index and task_index), and field columns (e.g., frame_index, state_0 to state_6, action_0 to action_6, etc.), with vector columns flattened and stored as single-precision floating-point numbers. This dataset is suitable for robot control, time series analysis, and machine learning tasks; users can read the data using the Apache TsFile SDK and filter specific episodes via SQL-like queries.